NextFin News - KKR has committed more than $3 billion to South Korea's artificial-intelligence supply chain in a pair of record-setting transactions, betting that the next bottleneck in the AI race is not chips but the power and facilities that run them. In July, the New York private-equity firm took management control of a 2 trillion won ($1.3 billion) renewable-energy platform with SK Inc., South Korea's largest clean-power venture. In August, it agreed to back SK Telecom's new AI data-center company, SK Horizon, in a 3.08 trillion won ($2.23 billion) equity investment alongside the IMM Investment-Stonebridge consortium. Combined, the two deals are among the largest private-capital wagers ever placed on Korea's AI buildout, and they place KKR on both sides of the same constraint: the data centers where the compute sits, and the electrons that feed them.
The timing is deliberate. Global AI infrastructure spending is compounding faster than grid capacity can be permitted and built, and Korea sits at the intersection of the two tightest links in the chain: high-bandwidth memory production and data-center power. KKR's move is a wager that the market is still pricing AI as a semiconductor story when it is becoming, just as fast, an infrastructure story.
The Two Deals: Compute Needs Watts
The first transaction targets the facilities. SK Telecom announced on Aug. 27 that it is splitting its wholly owned SK Broadband subsidiary into two entities. The surviving company keeps wired, media and enterprise operations; the new company, SK Horizon, houses the group's data-center and submarine-cable assets. KKR and the IMM consortium are injecting a combined 3.08 trillion won. Once the deal closes, KKR will own 29%, IMM 20%, and SK Telecom 51%, retaining management control. SK Horizon will operate 318 megawatts of AI data-center capacity across eight existing facilities — Seocho, Ilsan, Bundang, Gasan, Centum, Yangju and Pangyo — plus projects under construction in Ulsan and Guro. The spin-off is expected to complete in the first quarter of 2027, pending shareholder and regulatory approvals.
The second transaction targets the power. On July 1, KKR and SK Inc. launched what they described as South Korea's largest renewable-energy platform, valued at 2 trillion won. KKR takes initial management control; SK participates as an equity investor and retains an option to seek control rights through future talks. The platform consolidates wind, solar and fuel-cell assets from SK Innovation, SK ecoplant and SK eternix, starting with 1.7 gigawatts of operating capacity and scaling to 10 gigawatts — enough, the companies said, to power 100 large-scale 100-megawatt data centers simultaneously.
Read together, the pairing is the point. A single 100-megawatt data center can consume as much electricity as roughly 70,000 homes, and AI workloads run at high utilization around the clock, which moves power from a line item to the dominant operating expense. A developer that cannot guarantee reliable, preferably clean, power cannot sign a hyperscaler contract. Hyperscalers increasingly require carbon-free energy in their procurement standards, which is why a renewables platform is not an ESG accessory here — it is a commercial prerequisite. By securing both the load and the generation, KKR is underwriting the entire economics of Korea's AI infrastructure rather than one slice of it.
The Bottleneck Has Moved From GPUs to Grids
For the first wave of the AI boom, the scarce asset was the graphics processor. Nvidia, Samsung Electronics and SK Hynix raced to pack more compute onto each chip, and the market rewarded whoever could ship the most advanced silicon. That scarcity has not disappeared, but a second, slower-moving constraint has emerged: a data center cannot run without a grid connection, a substation, and a long-term power purchase agreement.
The transmission mechanism is physical, not financial. Training and inference workloads run 24 hours a day. Power costs therefore scale directly with utilization, and grid interconnection queues in most developed markets now stretch into years. That turns transmission capacity and generation assets into the gating input for the entire AI supply chain. Seoul underscored the strategic stakes when it announced three massive investment projects spanning semiconductors, physical AI and AI data centers — a signal that the state sees power and facilities as national infrastructure, not a private-sector afterthought.
KKR's Korea positioning mirrors moves on other continents. The firm bought the $4.2 billion North American arm of EDF Power Solutions to build renewables for US data centers, and acquired a 50% stake in a 1.5-gigawatt US solar portfolio from TotalEnergies for $1.25 billion. The same playbook — pair the compute load with the clean electrons — is being executed in parallel across the Pacific. The difference in Korea is scale relative to the market: a $1.3 billion renewables platform and a $2.23 billion data-center equity injection are large enough to move the domestic power market, not just participate in it.
Why Korea Makes Structural Sense
Three advantages make the bet rational rather than speculative. First, geography and latency: Korea sits in the same time zone as Japan and China and connects to both through undersea cables. SK Horizon will expand submarine-cable assets alongside its data centers, giving it a route to serve regional AI demand without the latency penalty of routing through the US West Coast. For inference workloads serving Asian users, that latency differential is a competitive advantage, not a rounding error.
Second, an existing industrial base. Samsung Electronics and SK Hynix already dominate high-bandwidth memory, the specialized DRAM that AI accelerators require. HBM is the highest-value link in the AI supply chain, and both companies are Korean. A data-center ecosystem anchored in the same country as the memory supply chain reduces logistics friction and aligns incentives across the chain. When SK Hynix ships HBM3E to a hyperscaler, the racks that consume it are increasingly likely to sit in the same jurisdiction that produced the memory — shortening the physical supply chain and reducing exposure to cross-border disruption.
Third, policy tailwinds. Seoul has made AI infrastructure a national priority, and the president's office has personally attended data-center ceremonies with SK Group and AWS executives. When a government treats data centers as strategic infrastructure, permitting, grid access and land become easier to obtain — precisely what a capital-intensive, multi-year buildout needs. Korea's power market is also being restructured to accommodate large industrial loads, which improves the odds that a 10-gigawatt renewables target is reachable rather than aspirational.
KKR is not acting alone. Brookfield Asset Management and other global managers are already scouting Korean AI data centers, and the firm's broader AI-infrastructure push includes a partnership with a group of six asset managers, including a tie-up with Nvidia, to mobilize more than $500 billion of third-party capital for AI infrastructure over time. KKR also structured or syndicated more than $80 billion of private investment-grade transactions in the first half of 2026 alone, in a company presentation — dry powder that needs deployment, and AI infrastructure is the designated landing zone.
Cyclical Bet or Structural Shift? The Judgment
Here is the call this piece must defend: the power-and-facilities layer of the AI supply chain is a structural shift, not a cyclical bet. The distinction matters because it determines whether KKR's capital earns a durable return or gets caught in a mean-reverting downturn.
A cyclical read would argue that data-center demand is front-loaded — hyperscalers are digesting last year's GPU purchases, utilization rates will fall, and new facilities will sit half-empty. That view has a pedigree. Several Wall Street analysts have warned that AI capital expenditure is running ahead of revenue, and that a pullback in 2027 is plausible. If they are right, contracted power assets still carry fixed costs, and returns on a 10-gigawatt buildout assumption would collapse.
The structural case is stronger, for three reasons. First, the constraint is physical, not financial. A GPU can be ordered and delivered within quarters; a 100-megawatt data center with grid interconnection and permitting takes years. Scarcity at the physical layer persists even when chip demand wobbles. Second, power demand from AI is compounding, not one-off. Inference traffic grows with adoption, and every new AI application adds load. Third, the asset base is not a greenfield gamble: SK Horizon's eight existing facilities already generate cash flow, and the 318-megawatt platform is partly operational before the first new tower goes up.
There is also a financial-engineering reason to side with the structural view. Infrastructure assets of this kind are typically contracted under long-term power purchase agreements with creditworthy counterparties, producing bond-like cash flows that are largely insulated from the GPU cycle. The cyclical risk sits in the GPU layer — where Nvidia's quarterly orders can swing with hyperscaler capex — while the structural demand sits in the power-and-facility layer. KKR has deliberately positioned itself in the structural layer.
The Second-Order Trade Everyone Is Missing
The first-order read of these deals is simple: KKR likes AI data centers. The second-order read is more interesting. KKR is not really betting on AI demand at all — it is betting on the convergence of two permissioned networks: the electric grid and the telecom cable system.
Data centers need grid interconnection. Submarine cables need landing rights and telecom licenses. Both are slow-to-build, locally monopolistic assets. Once you own them, competitors cannot easily duplicate them, which gives the owner a form of pricing power that a pure-play data-center operator — which can be undercut by a rival that finds cheaper land and power — does not have. In economic terms, KKR is buying assets with high barriers to entry and low marginal cost of replication for the owner, which is the classic definition of a durable moat.
The third-order implication follows: the winners in the next phase of the AI buildout may not be the most visible names. Investors chasing Nvidia and SK Hynix are buying the supply that everyone already knows is scarce. The less obvious scarcity is sitting in permitted land, grid queues, and cable landing stations — assets that private capital like KKR's is quietly accumulating while the market looks the other way. If that thesis holds, the highest returns in the AI supply chain over the next five years will accrue not to the chip designers but to the owners of the physical bottlenecks that chip designers depend on.
The Counter-Thesis, and What Would Break It
The strongest argument against this trade is that AI capital expenditure is a bubble, and when hyperscalers cut spending, the entire power-and-facility chain re-prices. This view is backed by mainstream analysts who have flagged that AI infrastructure spending may be outpacing monetization. It is not a strawman, and it deserves weight: if data-center utilization falls, a 10-gigawatt assumption becomes a liability rather than an option, and contracted power assets can still leave buyers negotiating harder on renewal.
There is also a Korea-specific risk that the counter-thesis leans on: the country's power grid is already tight, and a rapid data-center buildout could strain generation capacity, forcing curtailment or delaying interconnection. If the grid cannot deliver the watts that SK Horizon's facilities are built to consume, the assets underperform regardless of AI demand.
The falsifying signal is specific and observable: if hyperscaler AI-related capital expenditure growth falls below 10% year-over-year for two consecutive quarters while data-center vacancy rates in Korea rise above 15%, the structural thesis is wrong, and this trade is a cyclical bet dressed as a regime shift. The quarterly capex disclosures from the major cloud providers, combined with Korea's data-center vacancy data, are the two dials to watch. A secondary signal would be repeated interconnection delays or curtailment orders from Korea's grid operator — that would confirm the grid-constraint version of the bear case.
What Comes Next
Who benefits is clear: KKR's limited partners, SK Group's listed affiliates — SK Inc., SK Telecom, SK ecoplant and SK eternix — and the Korean construction and electrical-equipment firms that will build out the facilities. Who is exposed is equally clear: pure-play data-center operators without their own power assets, and any developer that assumed cheap, abundant grid capacity would remain available.
In the short term, over the next six to twelve months, expect more announcements as global private capital follows KKR into Korea. Volatility will be high because deal terms and regulatory approvals take time, and the SK Horizon spin-off does not close until the first quarter of 2027. KKR shares have traded in a wide range this year — opening near $100 and moving between roughly $83 and $165 over the past two years — reflecting broader uncertainty in the alternative-asset sector as much as anything specific to these deals.
In the medium term, over one to three years, execution risk is real. Permitting, grid interconnection, and equipment supply can all slip, and the renewables platform must scale from 1.7 gigawatts toward 10 gigawatts without losing its contracted power buyers. The 100-data-center equivalence is a useful benchmark: if KKR and SK cannot line up anchor tenants for even a fraction of that capacity, the 10-gigawatt target becomes a stranded-asset risk.
In the long term, over five years and beyond, if AI adoption compounds as KKR assumes, the power-and-facility layer becomes the highest-return segment of the AI supply chain, because it is the hardest to replicate.
The base case is that Korea becomes a regional AI infrastructure hub and KKR's two deals earn mid-teens returns as contracted cash flows compound. The upside case is that AI inference demand grows faster than expected, power scarcity intensifies, and the 10-gigawatt renewables target accelerates. The downside case is that hyperscaler capex slows, vacancy rises, and the 10-gigawatt assumption proves optimistic.
"We are pleased to support SKT as it establishes SK Horizon and expands its AI data centre infrastructure platform in Korea," said Keith Kim, a partner at KKR. "SK Horizon brings together an established operating platform, capacity under development and a strong strategic partner. Korea's advanced digital ecosystem and growing demand for AI capacity provide a strong foundation for SK Horizon's next phase of growth."
KKR's bet is not that AI will win. It is that whoever wins AI will need Korea's power — and that the firm which owns the watts will collect rent from every competitor.
Explore more exclusive insights at nextfin.ai.
